I discovered a regression for PR#6986 a month ago:

http://www.netbsd.org/cgi-bin/query-pr-single.pl?number=6986

we have a regression, and with gcc4 coming into the tree, I figured it
was time to poke the port-i386 maintainers.

in a nutshell gcc is doing FANCY_MATH_387 (or its modern equivalent) and
generating floating point calls which aren't emulated in the kernel.
this is a showstopper for FPU-less 80386 machines.  (486SLC and 486SX
would be in trouble as well).

the easiest thing would be to drop support for these older machines, but
it could be something as simple as FANCY_MATH_387 sneaking back into our
in-tree GCC configuration.

-- 
  Aaron J. Grier | "Not your ordinary poofy goof." | agrier@poofygoof.com
              "silly brewer, saaz are for pils!"  --  virt